Fractional CFO for AI Startups: What It Owns

A fractional CFO at an AI startup owns the three finance jobs that cannot wait for a full-time hire: the compute forecast and gross margin, pricing and revenue recognition, and ARR quality. A full-time finance chief runs $250,000 to $450,000 or more all-in. Our CFO-level tiers start at $3,500 a month.¹

I work with founders in the four industries I go real deep in: technology (AI and SaaS), healthcare and biotech, pharma, and pharmacy. This piece is for the AI founder who has outgrown a bookkeeper and cannot yet justify a six figure hire. I am not a CPA and the firm performs no attest work, so your auditor decides what counts as audit grade, and tax filing goes to a tax partner we trust. Every company number below is made up (illustrative, not a client) unless I cite it.

Picture a Monday call with the lead investor on your Series A. She asks two questions: what is your gross margin today, and what does it do if usage doubles? Your bookkeeper can tell her what you paid your cloud provider last month. Nobody on your team can tell her what you will pay next quarter, or which customers cost you money. That gap is the job. By the end of this piece you will know which finance jobs to hand to a fractional CFO first, what each tier costs and when a full-time hire starts to make sense.

What I would hand a fractional CFO first at an AI startup

A fractional CFO is a part-time senior finance executive who owns forecasting, cash, pricing math, investor reporting and the raise for a company that is not ready to carry a full-time one. At an AI startup I would put three jobs at the top of the list, because each one moves a number an investor will test: the compute forecast and gross margin, pricing and revenue recognition, and ARR quality. Andreessen Horowitz describes the role as owning the P&L and the strategic roadmap, with CEOs who understand the terrain and CFOs who own the map.² The close, the 13 week cash flow and the board deck sit behind those three.

The full-time version arrives later than most founders think. Bessemer says that "SaaS companies usually think about bringing on a full-time finance hire between $5 million and $25 million of ARR."³ It puts a true CFO past $50 million of ARR.³ CRV sees the same ladder by funding stage: "Series B is where multi-entity structures and audit readiness start to need a full-time CFO."⁴ Before that, CRV says, a seed company can run on clean bookkeeping and a fractional CFO, and a Series A company usually needs a head of finance. Jason Lemkin of SaaStr says a strong VP of Finance is usually enough between $2 million and $20 million of ARR.⁵ My take is that those ladders were drawn for SaaS. An AI company reaches the hard questions sooner, because its costs scale with usage from the first customer.

The compute forecast and your gross margin

The fractional CFO owns it, and the work starts by forecasting compute from usage, not as a percent of revenue. Here is the back of the napkin version, on illustrative numbers (not a client). An AI startup bills $250,000 a month on seat pricing. Its cloud and model bill is $110,000 and its other cost of revenue is $20,000, so cost of revenue is $130,000 and gross margin is 48%. Usage grows 40% in a quarter while the price stays flat. The compute bill becomes $154,000, cost of revenue becomes $174,000 and gross margin falls to 30.4%, with no change in revenue. A bookkeeper records that after the fact. A forecast shows it a quarter early, and that is the slide I would want in front of the board before an investor asks for it.

Plan on being compared to a moving target. ICONIQ's survey of about 300 software executives puts average AI gross margins at 41% in 2024 and 45% in 2025, with 52% projected for 2026,⁶ and its July edition projects 53% for 2026 and 59% for 2027.⁷ The spread is wide. Bessemer found that "On average, these AI Supernovas have only 25% gross margins"⁸ while its steadier group runs near 60%.

What I would watch is the slope, not the snapshot. CRV says Series A investors compare "current AI gross margin against the next 18 to 24 months of improvement"⁹ and look at the levers you control, such as inference and model routing. So the CFO's deliverable is a margin bridge: this quarter's margin, the specific changes that move it and the date each one lands. I would also never let compute sit in one expense line, because what serves customers belongs in cost of revenue and what trains models may not. I wrote the long versions in AI Gross Margin: Why Yours Isn't 80% and AI Compute Costs: COGS or R&D?.

Pricing and revenue recognition belong in the same meeting

The fractional CFO owns both, and they have to be decided together. Pricing is moving under founders' feet. The ICONIQ report text says "37% of companies plan to change their AI pricing model in the next year"¹⁰ though its chart for the same question reads 40% yes, 23% no and 37% unsure, so I read it as a direction, not a count. The same report has 58% of companies on a subscription or platform fee, 35% on usage-based pricing and 18% on outcome-based pricing. Every one of those choices has an accounting consequence, and that is the part founders tend to find out about last.

The consequence is timing. Deloitte's guidance on outcome-based pricing for AI agents frames the key judgment as whether you owe "a stand-ready obligation to provide continuous access to the agent"¹¹ or a specified quantity of successful outcomes, and that answer drives when revenue is recognized under ASC 606. EY makes the same point: whether you promised access or delivery changes how revenue is booked.¹² In my book that is why the CFO sits in the pricing meeting before the contract is signed, not after the first invoice. I cover the schedules in What Is Deferred Revenue in SaaS.

Two more rules belong on the CFO's calendar. FASB's ASU 2025-06 removes the stage-based rules for capitalizing internal-use software costs and uses a probable-to-complete threshold, effective for annual periods beginning after December 15, 2027.¹³ Your auditor will tell you what applies to a hosted or customer-facing product. On the tax side, Section 174A allows a deduction for domestic research costs paid or incurred in tax years beginning after December 31, 2024,¹⁴ while Cooley notes that foreign research costs are still amortized over 15 years.¹⁵ That matters if your engineers are offshore. Both questions go to your tax partner and your auditor. I flag them here so that someone owns them.

ARR quality, and what investors check before a raise

The fractional CFO owns the definition of ARR and the proof behind it. Investors know AI revenue can look recurring when it is not. CRV warns that "AI ARR often blends one-off contracts, credits-based pricing, performance-based arrangements and outcome-based revenue"¹⁶ Fortune reported that six VCs say founders count pilots, one-time deals or unactivated contracts as recurring revenue.¹⁷ TechCrunch quotes one VC who says some companies report contracted ARR about 70% higher than actual ARR, and the reporter's point is that CARR counts revenue before the product is live, so a slow rollout can end in a cancellation.¹⁸ Intel Capital calls usage that spikes and then drops to zero experimental run-rate revenue, neither annual nor recurring.¹⁹ I tell founders to keep it simple: show live ARR, contracted ARR and usage run-rate on separate lines and never add them together.

Then comes the data room. CRV tells Series A founders that bank statements should reconcile to their internal cash records, with any discrepancy investigated and documented before diligence starts, and it recommends a gross margin bridge in the metrics package.²⁰ Investors are also seeing a lot of AI decks. Carta reports that over 60 cents of each venture dollar on its platform went to AI companies in Q1 2026, and 83% of capital in SaaS went to AI startups.²¹ I would expect a crowded field to make diligence more careful, not less.

The worst version of an ARR problem is invented revenue, and there is a documented case. The SEC's June 2024 complaint against the founder of Joonko, an AI hiring startup, alleges she told investors the company had more than 100 customers and more than $1 million in revenue, that investors were defrauded of at least $21 million, and that she supplied falsified bank statements when an investor grew suspicious.²² Those are allegations in a civil case with no judgment that I found.²³ The Justice Department's September 11, 2026 release says she pleaded guilty to securities fraud, and says of the customers she named: "In truth and in fact, and as RAZ knew, these companies were never Joonko customers."²⁴ An earlier 2024 release counted about $27 million raised and noted that the charges were merely accusations at that stage.²⁵

This is a fraud case, not messy books, and I am not saying a fractional CFO would have stopped it. What I take from it is practical. The bank statement is the first document a careful investor asks for, and I want the person who reconciles it to be someone other than the person pitching the numbers. I learned how much finance an AI company needs by building a service line for it. At the national fractional firm where I worked from 2019 to 2026, I went from head of accounting to firm CFO and led the healthcare, biotech and technology vertical, and the job kept adding scope: and also this, and also that. That is where I learned that the bookkeeping at a technology company is the easy part and the judgment on top is the work, and it is why I went out on my own.

Where the bookkeeper stops and the CFO starts

Here is how I split the work. A bookkeeper keeps the transactions, a controller makes them reliable and a fractional CFO turns them into decisions. These are the rows that matter at an AI company, and the last column is the one most founders have nobody in.

When I would make the call

Sooner than a bookkeeper-only setup can survive, and Mercury lists the signs: a cash runway under 12 months with no clear plan to extend it, a fundraise within 6 months, a close that takes more than 15 business days or a forecast variance over 20%.²⁶ Brex's stage guide puts outsourced bookkeeping plus a part-time CFO at the seed stage and a head of finance at Series A.²⁷ My addition for AI companies is a fifth sign. When compute is one of your two largest cost lines and nobody forecasts it, you already have the job.

Here is how I match our tiers to stages, and it is my judgment, not a rule. Launch ($3,500 to $5,000 a month) is CFO-level foundational work, KPIs and monthly advisory, and it fits a seed-stage company building its first real forecast. Scale ($5,000 to $7,500) adds cash management, bill pay and a 13 week cash flow, which fits a company growing into a Series A. Exit ($10,000 a month plus a $10,000 one-time setup) is the full stack for the raise, the lender, diligence and the board room. Each tier includes everything below it, and the scope is flat and priced by the month, never by the hour.

What it costs next to a full-time hire

Annualized, Launch runs $42,000 to $60,000, Scale $60,000 to $90,000 and Exit $120,000 plus the $10,000 setup. A full-time CFO has a base salary of $215,000 to $326,250 on Robert Half's published range, which its page labels 2027,¹ and Workday, citing Robert Half, shows $195,500 to $321,750 for 2026.²⁸ Add bonus, benefits, payroll taxes, equity and recruiting, and a full-time finance chief commonly reaches $250,000 to $450,000 or more a year, higher in technology and high-cost markets. That all-in range is our estimate, not a survey result. Founderpath's small dataset puts CFO equity at Series A and B at 0.5% to 2.0%.²⁹

Set against the $250,000 to $450,000 range, Launch costs about 9% to 24% of the all-in figure, Scale about 13% to 36% and Exit's first year, with setup, about 29% to 52%. I would not read that as a bargain. I read it as buying the three jobs now and the full-time hire when the company is large enough to keep one busy. For the long version see How Much Does a Fractional CFO Cost in 2026? and When Should a Startup Hire a Fractional CFO.

If you run a biotech, pharma, pharmacy or AI company

For an AI company the three jobs above come first. For a biotech or pharma company the weight moves to accruals and the audit. Through a High Rock Accounting engagement, I served as the sole accountant on a clinical-stage biopharmaceutical client's month-end close, owning its preclinical, clinical, manufacturing and general accruals through its Nasdaq IPO. When a clinical vendor bills late, the estimate is your job. For a pharmacy, I would expect inventory and payer reconciliations to carry the weight, and that is my instinct from over a decade across two specialty pharmacies, not data. Our technology (AI and SaaS) page covers the founders who hit the three jobs above first.

If you only do 4 things after reading this, do these:

1. Name an owner for the compute forecast. Update it monthly from usage, and show gross margin by quarter with the changes that move it.
2. Write your ARR definition on one page. Live, contracted and usage run-rate, shown on separate lines.
3. Put revenue recognition in the pricing meeting. Decide how a price change books before you change the price.
4. Reconcile every bank account monthly. Keep the support, because the bank statement is the first thing a diligence team asks for.